review-code

Call get_workflow_command_guidance(kind="design-plan", focus="Architecture focus: inspect saved scripts against their current step objectives, input/output contracts, and relevant retained behavior. Propose improvements without applying them.

Load and apply read_skill(skills=[{"name":"builder-reference","path":"references/code-authoring.md"}]), then load and apply read_skill(skills=[{"name":"builder-reference","path":"references/scripted.md"}]) as review references; their authoring and execution instructions do not grant mutation or producing-run authority in this review.

Check source-layout migration consistency: resolve canonical code paths from workflow.json.code_layout_version. Version 1 uses code//main.py directly and shared imports rooted at WORKFLOW_CODE_ROOT, without execution-copy or copy-back assumptions. Absent/zero uses the supported legacy learnings//main.py layout. Inspect relevant active code, imports, and plan references for missing canonical files or stale layout assumptions; historical retained copies and an intentionally legacy layout are not defects. Do not migrate implicitly or infer the version from folder existence.

For a legacy scripted workflow, assess and normally propose migration to code/ using the Deliberate migration to code/ section of code-authoring.md. Include the complete source/helper/metadata scope, supported manifest-switch mechanism or its absence, rollback boundary, and focused runtime tests. Treat this as a maintenance improvement, not a bug; reuse existing proposals and respect explicit deferrals. If migration and testing are already authorized, preserve that scope for the implementation handoff rather than asking again. This review remains read-only.

For parameterized scripts, compare the declared script_parameters keys, types, required values, defaults, and enums with main.py parsing of STEP_PARAMS_JSON and its actual use of those values. Inspect relevant callers: execute_step supplies script_parameters; call_scripted_sub_agent supplies parameters. Check optional-value handling, hardcoded per-call values, undeclared inputs, and conflicting CLI/free-form parameter paths. Keep context_dependencies as positional sys.argv inputs, workflow configuration in VAR_*, and secrets out of the public parameter object. Parameterization is optional; recommend it only when repeated code rewriting or duplicated scripts shows a useful need.

Use current source and relevant retained evidence; do not launch a producing run merely to complete this review. Report material code/contract drift; skip unrelated operations checks and harmless recovered tool failures. Persist findings for a later /pulse-fixer; do not apply changes in this review.

{{context}}") and follow the returned instructions verbatim. Pass run_folder as the workflow's currently selected run folder when one is selected; otherwise omit it. Treat focus as the conversation/request context that appeared before the slash command, including the user's recent constraints and intent. The tool returns the canonical guided-flow text for this command — do not paraphrase or skip its steps.